modifying levels in factor variable using ifelse

Question

I wanted to modify levels in my factor variable by grouping two levels into one when I came across this strange situation. Basically, my new level is created, but all the remaining levels seem to be moved to the next one. Here is my example data, the code used and the output.

library(tidyverse) 
data <- structure(list(factor1 = structure(c(1L, 1L, 2L, 3L, 1L, 2L, 
        1L, 1L, 2L, 2L, 2L, 2L, 2L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
        1L, 1L, 1L, 3L, 1L, 1L, 1L, 4L), .Label = c("0", "1", "2", "3", 
        "4", "5", "6", "7"), class = "factor")), row.names = c(NA, -30L
        ), class = c("tbl_df", "tbl", "data.frame"), .Names = "factor1")
data_out <- data %>% mutate(factor1 = ifelse(factor1 %in% c('0', '1'), 
                                             factor1, '>1'))
structure(list(factor1 = c("1", "1", "2", ">1", "1", "2", "1", 
"1", "2", "2", "2", "2", "2", "1", "2", "1", "1", "1", "1", "1", 
"1", "1", "1", "1", "1", ">1", "1", "1", "1", ">1")), .Names = "factor1", 
class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA, -30L))

Is it desirable behaviour? It certainly isn't in my case. How could it be explained and then corrected?

Answer 1

I'm guessing this problem revolves around the way factors are constructed. How a factor goes from having levels of {"0", "1"} to levels {"1","2", ">1"} by way of mutate was still not clear to me.

R factors are actually base-1 integer vectors with attributes that are their levels. So your "0" levels initially were actually integer-1's and your "1" levels were integer-2's. Apparently the mutate function saw fit to create a new factor with an additional level that was printed as ">1" but also reassigned the "0" level to a new "1"-level and the "1" level to a "2"-level. This looks like a dangerous behavior on hte part of mutate to me. I think it should have given you either a new factor with levels "0","1",">1" or it should have thrown an error.

The error comes from ifelse although mutate compunds the problem by making the new column into a factor as well. If you coerce data to a dataframe, then you see:

data$factor2 <- ifelse( data$factor1 %in% c('0', '1'), 
                                              data$factor1, '>1')
data
#-------- same issue except
   factor1 factor2
1        0       1
2        0       1
3        1       2
4        2      >1
.... delete the other 26 rows
> str(data)
'data.frame':   30 obs. of  2 variables:
 $ factor1: Factor w/ 8 levels "0","1","2","3",..: 1 1 2 3 1 2 1 1 2 2 ...
 $ factor2: chr  "1" "1" "2" ">1" ...

This would have let you stay in the dplyr package:

recode_factor(data$factor1, `0` = "0", `1` = "1", .default=">1")
 [1] 0  0  1  >1 0  1  0  0  1  1  1  1  1  0  1  0  0  0  0  0  0  0  0  0  0  >1 0  0  0  >1
Levels: 0 1 >1

Answer 2

Just in case of someone struggling with similar issue in future and looking for a easy way to group these factors without reassigned remaining one:

fct_collapse(data$factor1, '>1' = c('2', '3'))